Very High Embedding Capacity Algorithm for Reversible Image Watermarking

Reversible image watermarking enables the embedding of copyright or useful information in a host image without any loss of information. Here a novel technique to improve the embedding capacity i.e. reversible watermarking using an adaptive prediction error expansion & pixel selection is proposed. This work is an improvement in conventional Prediction Error Expansion by adding two new techniques adaptive embedding & pixel selection. Instead of uniform embedding, here one or two bits of watermark are adaptively embed into the expandable pixels as per the regional complexity. Adaptive Prediction Error Expansion can obtain the embedded rate upto 1.3 bits per pixel as compared to the 1 BPP of conventional Prediction Error Expansion. Also an intermediate step of prediction error expansion is proposed to select relatively smooth pixels and ignore the rough ones. In other words, the rough pixels may remain unchanged, and only smooth pixels are expanded or shifted. Therefore compared with conventional Prediction Error Expansion, a more sharply distributed prediction error histogram is obtained i.e. , and a larger proportion of prediction-errors in the histogram are expanded to carry hidden data. So the amount of shifted pixels is diminished, which leads to a better image quality. With these improvements, this method performs better than conventional Prediction Error Expansion. It can embed larger payloads with less distortion (almost 30% greater than the conventional method). KeywordsReversible image watermarking, Adaptive prediction error expansion, Gradient adjusted prediction, Pixel Selection (PS). __________________________________________________*****_________________________________________________